Improving Neural Network Efficiency via Post-training Quantization with Adaptive Floating-Point | IEEE Conference Publication | IEEE Xplore